D4ASCR: Dataset for Aluminum Strip Cold Rolling

This is the dataset of our work: Controlling Aluminum Strip Thickness by Clustered Reinforcement Learning with Real-world Dataset.

Our dataset D4ASCR (Dataset for Aluminum Strip Cold Rolling) is constructed based on historical rolling data from a real-world aluminum strip rolling line, incorporating a safety-constrained Smith predictor. Aluminum strip cold rolling trajectories are collected from a real-world production line. The logging policy, based on expert knowledge utilizing the Smith predictor, is employed to construct D4ASCR. The interaction data encompasses all process parameters and various production conditions, including actions, states, and rewards calculated based on the RL reward function.